What do you need to run?
Describe the outcome once. Fabric searches compatible capacity across the market.
Workload brief
ComfyUI image pipeline
Your requirement, ready for a market search.
Searching the compute market
Checking the live market snapshot. The next screen illustrates the proposed routing workflow; its example routes are not live quotes.
Choose your compute.
Compare specifications and estimated compute cost. Availability, startup time and performance require confirmation before a real launch.
Ready for your workload.
These settings illustrate the information needed for a real deployment. No keys or files are sent to an infrastructure venue.
Ephemeral files are deleted at termination. Persistent storage and network charges require a separate quote. Only a public SSH key would be requested for a real launch.
A clear launch checklist.
Exact GPU variant and quantity · region · image compatibility · storage lifecycle · access · spending limit.
For multi-GPU jobs, interconnect and topology must be checked separately. Memory across cards is not automatically a single shared pool.
Review your workspace.
Review the configuration and infrastructure details before continuing the simulation.
Fabric is launching your route.
Fabric is coordinating the provider, environment and health check underneath.
Your GPU workspace is ready.
This simulated workspace shows the planned access and management experience. No real GPU has been provisioned.
Instance IDStarted 6 minutes ago
Reservation expires in 7h 54mVast.ai · Netherlands
Underlying route · illustrative
Access your machine
⌘ Web terminal
Use the machine directly inside Fabric.
SSH
Demo address · illustrative only
ssh fabric@fc-demo.fabric.compute -p 22417ComfyUI
HEALTHY
Routed application service
JupyterLab
AVAILABLE
Notebook environment
Connection & environment details
All public-demo values below are illustrative. No secrets, keys or passwords are displayed.
Know what survives termination.
Workspace disk
200 GB · persists while reservation is active
Checkpoint locationfabric://fc-demo-24091/checkpoints
Local temporary storage
Deleted when the instance is terminated
$0.09 accrued
Machine and workload performance
Delivered in 72 seconds
Recent logs STREAMING · DEMO
16:40:07 health check passed · all services responding
16:41:02 model loaded to cuda:0 · 31.4 GB
16:41:07 worker ready · queue listening
16:44:18 batch 0042 completed · 20.7 images/minCheaper equivalent capacity found
Fabric found a compatible route that could save $1.19 at the next safe checkpoint.
Every workload, one control plane.
Follow launches, performance and cost without switching provider accounts.
The market Fabric routes through.
Observed provider offers from the connected market feeds. Availability and price can change before an order is placed.
H100 SXM/HGX 80GB · observed direct inventory
Requesting the latest normalized availability snapshot…
Web terminal.
Operate routed compute without leaving Fabric or managing another provider account.
Try: nvidia-smi · fabric status · ls · pwd · clear
One RFQ reaches the market.
Compare executable single-cluster and split-fill routes on identical terms.
New capacity request
Fabric handles the market underneath.
Your requirement is normalised once, then compared across direct providers, aggregators and private capacity partners.
12 providers checked. 4 executable. 1 Fabric price.
Every provider, one account.
Track compute cost, Fabric fees and savings against the market benchmark.